An unmanned aerial vehicle remote sensing and multi-source data fusion earthquake comprehensive emergency thematic map generation method

Abstract

The application relates to an unmanned aerial vehicle remote sensing and multi-source data fusion earthquake comprehensive emergency special map generation method. The method comprises the following steps: unifying the time and space references of obtained unmanned aerial vehicle remote sensing images, satellite images, ground sensor data and rescue report text data, and generating a unified time and space data cube; cross-modal interaction fusion is performed on visual data features and semantic data features in the cube to generate a cross-modal collaborative feature matrix; local conflict-based evidence theory fusion is performed on the matrix to generate a comprehensive probability mass function; multi-dimensional quantitative evaluation indexes are generated by combining disaster chain coupling effect cascading risk quantitative evaluation on the function; and the indexes are dynamically synthesized based on a disaster chain logic ontology and visual saliency to generate an earthquake comprehensive emergency special map. The method can eliminate multi-source data fusion conflicts and realize multi-dimensional comprehensive disaster condition evaluation by combining disaster chain coupling effects.

Description

Technical Field

[0001] This invention belongs to the field of earthquake emergency rescue technology, and in particular relates to a method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data. Background Technology

[0002] With the continuous development of earthquake emergency rescue technology, rapid disaster perception technologies such as UAV remote sensing have emerged. This technology is characterized by its mobility and flexibility, as well as its ability to acquire high-resolution images of disaster areas. This has led to the current method of generating earthquake emergency thematic maps based on remote sensing data.

[0003] Traditional technologies typically utilize drones or satellites to acquire images of disaster areas. Visual interpretation of these images is then used to extract surface disaster information, such as building damage, and to generate emergency thematic maps. For multi-source data, such as ground sensor monitoring data or rescue report texts, these data are often analyzed independently and then simply overlaid or annotated on the thematic map. This lacks in-depth correlation and interaction of heterogeneous data at the feature level.

[0004] However, current traditional methods have limitations in effectively integrating multi-source heterogeneous data to eliminate conflicts and uncertainties in disaster assessment, and lack comprehensive consideration of the coupling effect of disaster chains. This results in the generated emergency thematic maps having a single dimension for disaster assessment and failing to comprehensively and accurately reflect the overall disaster situation. Summary of the Invention

[0005] Based on this, it is necessary to provide a method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, which can eliminate conflicts in the fusion of multi-source heterogeneous data and combine the disaster chain coupling effect to achieve multi-dimensional comprehensive disaster assessment.

[0006] Firstly, this application provides a method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, including:

[0007] S1. Acquire UAV remote sensing images, satellite images, ground sensor data and rescue report text data, and unify the spatiotemporal reference of UAV remote sensing images, satellite images, ground sensor data and rescue report text data to generate a unified spatiotemporal data cube.

[0008] S2. Perform cross-modal interactive fusion of visual data features and semantic data features in the unified spatiotemporal data cube to generate a cross-modal collaborative feature matrix;

[0009] S3. Perform evidence theory fusion based on local conflict allocation on the cross-modal collaborative feature matrix to generate a comprehensive probability quality function;

[0010] S4. Conduct a cascaded risk quantification assessment of the comprehensive probability quality function, combining the disaster chain coupling effect, and generate multi-dimensional quantitative assessment indicators.

[0011] S5. Dynamically synthesize multi-dimensional quantitative assessment indicators based on disaster chain logical ontology and visual saliency to generate a comprehensive earthquake emergency thematic map.

[0012] In one embodiment, S2 includes:

[0013] S21. Extract edge features from the UAV remote sensing images in the unified spatiotemporal data cube to generate a UAV edge feature map;

[0014] S22. Based on the edge feature map of the UAV, guide filtering feature sharing is performed on the low-level features of satellite images in the unified spatiotemporal data cube to generate a visual fusion feature map;

[0015] S23. Encode and map the physical quantity statistical features of ground sensor data and the word vector sequence of rescue reporting text data in the unified spatiotemporal data cube to generate semantic feature vectors;

[0016] S24. Use the visual fusion feature map as the query matrix and the semantic feature vector as the key matrix and value matrix to perform cross-attention calculation to generate a cross-modal attention score matrix;

[0017] S25. Based on the cross-modal attention score matrix, perform weighted fusion mapping on the visual fusion feature map and semantic feature vector to generate a cross-modal collaborative feature matrix, which contains visual feature components and semantic feature components.

[0018] In one embodiment, S3 includes:

[0019] S31. Map the visual feature components and semantic feature components to the identification frameworks for personnel casualty status, infrastructure damage status and hazardous chemical risk status, respectively, and perform basic probability allocation transformation to generate basic probability allocation functions for visual evidence and basic probability allocation functions for semantic evidence.

[0020] S32. Calculate the evidence distance conflict metric for the basic probability assignment functions of visual evidence and semantic evidence, and generate a distance metric value. The distance metric value is calculated using the following formula:

[0021]

[0022] in, For distance metrics, and These are the basic probability assignment functions for visual evidence and semantic evidence, respectively. The Jaccard distance matrix;

[0023] S33. Perform a non-linear mapping of similarity and credibility weights on the distance metric to generate credibility allocation weights;

[0024] S34. Based on the credibility allocation weight, perform local conflict reassignment on the basic probability allocation function of visual evidence and the basic probability allocation function of semantic evidence to generate the corrected visual evidence probability function and the corrected semantic evidence probability function.

[0025] S35. Perform orthogonal summation on the modified visual evidence probability function and the modified semantic evidence probability function to generate a comprehensive probability quality function.

[0026] In one embodiment, S4 includes:

[0027] S41. Obtain the pre-earthquake building vulnerability index, and perform structural failure risk quantification calculation on the probability of infrastructure damage state in the comprehensive probability quality function and the pre-earthquake building vulnerability index to generate a structural damage index.

[0028] S42. Obtain meteorological and environmental parameters, perform quantitative calculations of hazardous chemical diffusion risk based on the probability of hazardous chemical risk state in the comprehensive probability mass function and meteorological and environmental parameters, and generate a hazardous chemical risk index.

[0029] S43. Extract the population time series baseline and evacuation time series rate from the unified spatiotemporal data cube. Based on the structural damage index and hazardous chemical risk index, perform risk quantification calculations on the trapped personnel baseline and damage amplification effect on the population time series baseline and evacuation time series rate to generate the casualty index. The formula for calculating the casualty index is as follows:

[0030]

[0031] in, The casualty index, for Population time series baseline at any given moment. for evacuation timing rate for Time-based structural damage index This is the damage amplification factor for hazardous chemicals to personnel. for Hazardous chemical risk index at all times;

[0032] S44, combined structure damage index, hazardous chemical risk index and personnel casualty index, generate multi-dimensional quantitative assessment indicators.

[0033] In one embodiment, S42 includes:

[0034] S421. Based on the probability of hazardous chemical risk status, perform abnormal smoke region segmentation on the visual data in the unified spatiotemporal data cube to generate abnormal smoke contour boundaries.

[0035] S422. Replace the abnormal smoke contour boundary with the default diffusion variance of the spatial concentration diffusion model to generate a dynamically corrected diffusion variance.

[0036] S423. Obtain the ground gas sensor concentration, perform leakage source strength inversion calculation on the dynamic correction diffusion variance and the ground gas sensor concentration, and generate dynamic leakage source strength.

[0037] S424. Obtain wind speed and direction from meteorological environmental parameters. Based on dynamic leakage source strength and dynamic correction diffusion variance, perform integral calculations on wind speed, wind direction, and spatial concentration parameters to generate a hazardous chemical risk index. The formula for calculating the hazardous chemical risk index is as follows:

[0038]

[0039] in, Indicates the risk index of hazardous chemicals. For dynamic leakage source strength, For wind speed, For the height of the leak, and To dynamically correct the components of diffusion variance in the horizontal and vertical directions, and For spatial coordinate variables, Let be the integral volume space.

[0040] In one embodiment, S5 includes:

[0041] S51. Classify the structural damage index, casualty index, and hazardous chemical risk index based on the natural breakpoints of the disaster chain ontology logic constraints, and generate logically consistent classification evaluation indicators.

[0042] S52. Dynamically symbolize and render logically consistent hierarchical evaluation indicators to generate personnel casualty isosurface layers, hazardous chemical flow particle layers, and infrastructure damage base map layers.

[0043] S53. Based on the intensity of disaster indicators, dynamic transparency channel values ​​are assigned to the personnel casualty isosurface layer, hazardous chemical flow particle layer, and infrastructure damage base map layer for adaptive synthesis to generate an initial synthesized thematic map. The formula for calculating the fused pixel value of the initial synthesized thematic map is as follows:

[0044]

[0045] in, These are the output pixel values ​​of the initial composite thematic map. and The dynamic transparency of the personnel casualty layer and the hazardous materials layer are respectively. , and These are the original pixel values ​​corresponding to the personnel casualty layer, the hazardous materials layer, and the infrastructure layer, respectively.

[0046] S54. Calculate the inner product of structural damage gradient and hazardous chemical risk gradient for the spatial distribution of infrastructure layer and hazardous chemical layer in the initial synthetic thematic map, and generate the indicator profile of active disaster chain area.

[0047] S55. Overlay the active zone indicator outline of the disaster chain onto the initial synthetic thematic map to generate a comprehensive earthquake emergency thematic map.

[0048] In one embodiment, S54 includes:

[0049] S541. Perform spatial differential operator convolution on the pixel spatial distribution of the infrastructure layer and the hazardous chemicals layer in the initial synthesized thematic map to generate the spatial gradient field of structural damage and the spatial gradient field of hazardous chemicals risk.

[0050] S542. Perform pixel-by-pixel multiplication and summation operations on the spatial gradient field of structural damage and the spatial gradient field of hazardous chemical risk to generate the gradient inner product field.

[0051] S543. Binarize the gradient inner product field with a set threshold and extract the edge to generate a disaster chain active area indicator profile.

[0052] Secondly, this application also provides a device for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, including:

[0053] The spatiotemporal unification module is used to acquire UAV remote sensing images, satellite images, ground sensor data and rescue report text data, and to unify the spatiotemporal reference of UAV remote sensing images, satellite images, ground sensor data and rescue report text data to generate a unified spatiotemporal data cube.

[0054] The cross-modal fusion module is used to perform cross-modal interactive fusion of visual data features and semantic data features in a unified spatiotemporal data cube to generate a cross-modal collaborative feature matrix.

[0055] The evidence fusion module is used to perform evidence theory fusion on cross-modal collaborative feature matrices based on local conflict allocation to generate a comprehensive probability quality function.

[0056] The cascaded assessment module is used to conduct a cascaded risk quantification assessment of the comprehensive probability quality function in combination with the disaster chain coupling effect, and generate multi-dimensional quantitative assessment indicators.

[0057] The thematic map generation module is used to dynamically synthesize multi-dimensional quantitative assessment indicators based on the disaster chain logical ontology and visual saliency to generate a comprehensive earthquake emergency thematic map.

[0058] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the earthquake integrated emergency thematic map generation method of UAV remote sensing and multi-source data fusion as described in the first aspect.

[0059] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the earthquake integrated emergency thematic map generation method as described in the first aspect, which integrates UAV remote sensing and multi-source data.

[0060] The aforementioned method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data constructs a unified spatiotemporal data cube by unifying the spatiotemporal references of heterogeneous remote sensing images, sensor data, and text data from multiple sources. It then performs cross-modal interactive fusion of visual and semantic features to generate a collaborative feature matrix. Furthermore, it utilizes evidence theory based on local conflict allocation to fuse the collaborative feature matrix, eliminating conflicts and uncertainties between heterogeneous data and generating a comprehensive probability quality function. Subsequently, it combines the disaster chain coupling effect to perform cascaded risk quantification assessment of the comprehensive probability quality function, outputting multi-dimensional quantitative assessment indicators. Finally, it dynamically synthesizes the assessment indicators based on the disaster chain logical ontology and visual saliency, thereby achieving deep collaboration and conflict resolution of multi-source heterogeneous disaster data. This effectively overcomes the limitations of a single assessment dimension and achieves the technical effect of comprehensively and accurately reflecting the cascaded impact of disaster chains and the overall disaster situation. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart illustrating a method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, provided by this invention.

[0063] Figure 2 This is a schematic diagram of the structure of an earthquake integrated emergency thematic map generation device that integrates UAV remote sensing and multi-source data, provided by the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] In one embodiment, such as Figure 1 As shown, a method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data is provided. This embodiment illustrates the application of this method to an earthquake emergency processing terminal. It is understood that this method can also be applied to an earthquake emergency processing server, and further to a system including both an earthquake emergency processing terminal and an earthquake emergency processing server, and is implemented through the interaction between the two. In this embodiment, the method includes the following steps:

[0066] S1. Acquire UAV remote sensing images, satellite images, ground sensor data, and rescue report text data, and unify the spatiotemporal reference of UAV remote sensing images, satellite images, ground sensor data, and rescue report text data to generate a unified spatiotemporal data cube.

[0067] Optionally, the earthquake emergency response terminal can capture four types of raw disaster data: UAV remote sensing imagery, satellite imagery, ground sensor data, and rescue report text data. UAV remote sensing imagery consists of real-world images of the disaster area collected from low-altitude flights; satellite imagery consists of large-scale surface images transmitted from high-altitude remote sensing satellites; ground sensor data consists of geological and environmental sensing monitoring data collected by various sensing devices deployed within the disaster area; and rescue report text data consists of text-based disaster records provided by rescue personnel.

[0068] Specifically, the earthquake emergency processing terminal performs unified spatiotemporal reference processing based on preset spatial coordinate rules and timestamp calibration rules. At the spatial level, a unified geodetic coordinate system is used to complete the geometric registration and correction of all image data. At the temporal level, time markers are added to various types of data according to a unified time scale.

[0069] Preferably, the earthquake emergency processing terminal will arrange the multi-type data that have completed the benchmark unification in a structured manner according to the spatial dimension, temporal dimension, and data attribute dimension to construct a data cube (spatiotemporal data cube). This three-dimensional data structure can realize the orderly storage of multi-source heterogeneous data and provide a standardized data carrier for subsequent feature extraction and fusion.

[0070] S2. Perform cross-modal interactive fusion of visual data features and semantic data features in the unified spatiotemporal data cube to generate a cross-modal collaborative feature matrix.

[0071] Optionally, the earthquake emergency processing terminal performs attribute partitioning on the data within the unified spatiotemporal data cube, resulting in visual data and semantic data. The visual data includes UAV remote sensing imagery and satellite imagery, while the semantic data includes ground sensor data and rescue reporting text data. Cross-modal data refers to data categories with different data types and different forms of expression; here, it refers to the heterogeneous data forms of image visual data and text sensor semantic data.

[0072] Specifically, the earthquake emergency response terminal extracts visual data features based on pixel operation rules. These visual data features include visual information such as the outlines of buildings on the ground, the texture of ground damage, and the boundaries of landform deformation. At the same time, it extracts semantic data features based on text segmentation and sensor numerical analysis rules. These semantic data features include geological stress values, text information on the distribution of affected people, and text information describing site damage.

[0073] Furthermore, the earthquake emergency response terminal establishes a feature interaction mapping relationship, completes the cross-correlation matching of two types of features, arranges the matched fused features in a row and column matrix format, and generates a cross-modal collaborative feature matrix. The internal elements of the cross-modal collaborative feature matrix are the uniformly quantized feature weight values.

[0074] S3. Perform evidence theory fusion based on local conflict allocation on the cross-modal collaborative feature matrix to generate a comprehensive probability quality function.

[0075] Optionally, the earthquake emergency response terminal retrieves the cross-modal collaborative feature matrix for fusion calculation. Data fusion processing is completed using Dempster-Shafer Theory (DST), a data analysis algorithm for handling uncertain information and fusing multiple pieces of evidence. Local conflict allocation is an optimized allocation rule for inconsistent judgment results and data conflicts among multi-source feature data. Specifically, the earthquake emergency response terminal first treats each feature element within the cross-modal collaborative feature matrix as independent disaster evidence, calculates the basic confidence level for each piece of evidence, and simultaneously calculates the conflict coefficient between different pieces of evidence. The conflict coefficient characterizes the degree of difference in judgment between multiple pieces of disaster evidence.

[0076] Furthermore, the earthquake emergency response terminal allocates conflict coefficients to various types of conflicting evidence according to local conflict allocation rules, corrects the confidence bias of individual pieces of evidence, and performs a superposition operation of all corrected evidence using an evidence theory fusion formula to obtain a comprehensive probability quality function. This comprehensive probability quality function characterizes the confidence probability of different disaster risks occurring. The corresponding evidence theory fusion formula is as follows:

[0077]

[0078] in, This represents the probability mass value after fusion. To address the issue of disaster identification, For the first The independent and heterogeneous original evidence corresponds to the first Disaster-related propositions The initial base probability mass assignment, The first one, representing the pre-defined framework for earthquake disaster identification. The category of independent disaster propositions, specifically encompassing propositions on casualties, infrastructure damage, and hazardous chemical risks, serves as the fundamental unit for determining the confidence level of disaster claims in evidence theory. This represents the conflict coefficient of evidence.

[0079] S4. Conduct a cascaded risk quantification assessment of the comprehensive probability quality function in conjunction with the disaster chain coupling effect to generate multi-dimensional quantitative assessment indicators.

[0080] Optionally, the earthquake emergency response terminal retrieves a comprehensive probability quality function to conduct risk quantification calculations. The disaster chain is a chain of disaster reactions that induce secondary and derivative disasters from the primary earthquake disaster. Common disaster coupling phenomena include earthquake-induced landslides, ground subsidence, and the formation of barrier lakes.

[0081] Specifically, the earthquake emergency response terminal pre-stores the logical relationships of disaster chains, sorts out the induction and transmission patterns between different disasters, calculates the disaster transmission risk layer by layer using a cascaded progressive deduction method, and characterizes the intensity of the correlation impact between different disasters based on the disaster coupling coefficient.

[0082] Furthermore, the earthquake emergency response terminal, combining the disaster confidence probability within the comprehensive probability quality function, calculates four types of assessment values: building damage risk, geological disaster risk, rescue access risk, and personnel injury risk. These four types of assessment values ​​are then integrated and standardized into data indicators, generating multi-dimensional quantitative assessment indicators. The corresponding risk assessment formula is:

[0083]

[0084] in the formula For the first Disaster risk assessment values, This is the evidence weighting coefficient. After multiple heterogeneous pieces of evidence undergo local conflict allocation and evidence fusion calculations, the overall comprehensive evidence corresponds to the [number]th [item]. Disaster-related propositions The final fusion probability quality value, For the first Class II disasters and the first Coupling coefficient of disaster-like events.

[0085] S5. Dynamically synthesize multi-dimensional quantitative assessment indicators based on disaster chain logical ontology and visual saliency to generate a comprehensive earthquake emergency thematic map.

[0086] Optionally, the earthquake emergency processing terminal completes layer synthesis processing based on multi-dimensional quantitative evaluation indicators. The ontology (ontology knowledge base) is a pre-constructed logical association knowledge base of earthquake disasters. This knowledge base stores the association logic rules of disaster categories, disaster impact range, and disaster level classification. Visual saliency refers to the characteristics of disaster areas in the image that are different from the surrounding areas and have higher visual recognition. Damaged buildings and fractured surfaces are all high visual saliency areas.

[0087] Specifically, the earthquake emergency response terminal retrieves the disaster chain logic ontology to filter and match the layer elements corresponding to various disaster situations. At the same time, it prioritizes the image layers according to their visual salience, with disaster layers with high visual salience being rendered first.

[0088] Furthermore, the earthquake emergency processing terminal dynamically adjusts the transparency, display range, and annotation style of different layers to complete the overlay and synthesis of disaster risk layers, geographic topography layers, and rescue point layers, ultimately generating a comprehensive earthquake emergency thematic map with complete disaster annotations, risk level classifications, and geographic location markers.

[0089] In the aforementioned method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, the earthquake emergency processing terminal eliminates disaster assessment biases in multi-source data through local conflict allocation rules. Simultaneously, it combines the disaster chain coupling effect to complete multi-level risk simulations, overcoming the shortcomings of traditional assessment methods with their single-dimensionality. The resulting comprehensive earthquake emergency thematic map comprehensively covers various disaster information such as building damage, geological risks, and rescue conditions. It boasts higher data accuracy and provides a clear and intuitive disaster display, offering reliable data support for earthquake emergency rescue deployment and disaster assessment analysis, and meeting the needs of emergency mapping in complex earthquake-stricken areas.

[0090] In one embodiment, S2 may include:

[0091] S21. Extract edge features from the UAV remote sensing images in the unified spatiotemporal data cube to generate UAV edge feature maps.

[0092] Optionally, the earthquake emergency processing terminal performs pixel-level grayscale gradient calculations on the UAV remote sensing images carried within the unified spatiotemporal data cube. Edge features refer to the contour boundary information formed by abrupt changes in pixel grayscale in the image, including concrete information such as building outlines, surface cracks, and boundaries of damaged areas.

[0093] Specifically, the earthquake emergency processing terminal uses a conventional gray-scale difference detection method to traverse all pixels in the UAV remote sensing image, calculates the gray-scale difference between adjacent pixels point by point, filters out pixels whose gray-scale difference exceeds the normal stable range, arranges the filtered pixels in a connected component regularization, retains effective contour lines and removes scattered noise pixels, and rearranges the pixel contour information according to the image coordinate correspondence to generate a complete and regular UAV edge feature map.

[0094] S22. Based on the edge feature map of the UAV, guide filtering feature sharing is performed on the low-level features of satellite images in the unified spatiotemporal data cube to generate a visual fusion feature map.

[0095] Optionally, guided filtering is an image processing method that constrains the filtering process of the image to be processed based on a guiding image. It can smooth redundant information while preserving edge contour details. The earthquake emergency processing terminal uses the UAV edge feature map as the guiding constraint benchmark and performs full-domain traversal processing on the bottom-level original features of satellite images in a unified spatiotemporal data cube. For example, based on the neighborhood pixel association rules of guided filtering, the earthquake emergency processing terminal synchronously shares the contour constraint information of the UAV edge feature map to the bottom-level features of the satellite image. It performs noise smoothing and feature scale alignment processing on the bottom-level features of the satellite image, so that the bottom-level features of the satellite image conform to the contour distribution pattern of the UAV edge feature map, completes the feature association integration across remote sensing images, and generates a visual fusion feature map.

[0096] S23. Encode and map the physical quantity statistical features of the ground sensor data and the word vector sequence of the rescue reporting text data in the unified spatiotemporal data cube to generate semantic feature vectors.

[0097] Optionally, word vectors are a form of expression that converts natural language text words into fixed-dimensional numerical vectors. The earthquake emergency processing terminal first performs multi-dimensional statistical summarization on the ground sensor data in the unified spatiotemporal data cube, extracting the physical quantity statistical features such as the mean, distribution range, and fluctuation trend of the physical quantities corresponding to the ground sensor data. At the same time, it performs word segmentation and semantic unit division on the rescue reporting text data in the unified spatiotemporal data cube, and converts the segmented semantic units one by one into standard word vector sequences.

[0098] Furthermore, the earthquake emergency processing terminal adopts a unified dimensional coding rule to map the statistical features of physical quantities and word vector sequences to the same numerical dimensional space, eliminating the heterogeneous dimensional differences between the two types of semantic data and generating semantic feature vectors with regular and unified dimensions.

[0099] S24. Use the visual fusion feature map as the query matrix and the semantic feature vector as the key matrix and value matrix to perform cross-attention calculation to generate a cross-modal attention score matrix.

[0100] Optionally, cross-attention is used to calculate the association matching weights between different types of features, enabling the retrieval of mutual associations between heterogeneous features. The earthquake emergency processing terminal performs tensor dimension normalization on the visual fusion feature map, converting the two-dimensional image feature tensor into a two-dimensional array structure and setting it as the query matrix. Simultaneously, it expands the semantic feature vectors to higher dimensions, constructing key and value matrices adapted to the query matrix dimensions. The earthquake emergency processing terminal calculates the matching association degree between the query matrix and the key matrix using the feature association degree calculation formula, then performs normalization constraint processing to quantify the association weights, fully preserving the pairwise association degree between visual and semantic features, and generating a cross-modal attention score matrix with a numerical distribution representing the association strength. The corresponding feature association degree calculation formula is as follows:

[0101]

[0102] In the formula, For the first Line 1 List attention score values, For querying the matrix, the first 3D feature vectors The key matrix is ​​the first 3D feature vectors This is the scaling factor for the feature dimension.

[0103] S25. Based on the cross-modal attention score matrix, perform weighted fusion mapping on the visual fusion feature map and semantic feature vector to generate a cross-modal collaborative feature matrix, which contains visual feature components and semantic feature components.

[0104] Optionally, the earthquake emergency processing terminal uses the value of the cross-modal attention score matrix as the benchmark weight for feature weighting, and fully reads all visual pixel feature information contained in the visual fusion feature map and all semantic dimension value information contained in the semantic feature vector.

[0105] Specifically, the earthquake emergency processing terminal performs dimensional weighted assignment operations on the visual fusion feature map and semantic feature vector according to the weights corresponding to the cross-modal attention score matrix. Then, it performs dimensional alignment and mapping splicing on the two types of features after weighted operations, splitting and dividing the independent visual feature components and semantic feature components. The visual feature components and semantic feature components are integrated into an overall array structure according to a fixed row and column arrangement format, generating a well-structured cross-modal collaborative feature matrix containing dual-component information.

[0106] In the above embodiments, the earthquake emergency processing terminal sequentially completes edge extraction of UAV remote sensing images, feature sharing of satellite image guidance filtering, encoding and mapping of multi-source semantic data, cross-modal cross-attention operation, and dual-feature weighted fusion, thereby achieving layer-by-layer interpretation, dimensional unification, and correlation interaction between remote sensing visual data and sensor text semantic data. This process, based on conventional image gradient operation, guidance constraint processing, and feature attention matching, effectively establishes the visual feature correlation between UAVs and satellite imagery, while simultaneously achieving standardized semantic expression of ground sensor data and rescue text data, accurately establishing the intrinsic connection between visual and semantic cross-modal features.

[0107] In one embodiment, S3 may include:

[0108] S31. Map the visual feature components and semantic feature components to the identification frameworks for personnel casualty status, infrastructure damage status, and hazardous chemical risk status, respectively, and perform basic probability allocation transformation to generate basic probability allocation functions for visual evidence and basic probability allocation functions for semantic evidence.

[0109] Optionally, the Basic Probability Assignment (BPA) is a numerical mapping relationship in evidence theory that assigns basic confidence values ​​to each proposition within the identification framework. The earthquake emergency response terminal retrieves the visual and semantic feature components contained in the cross-modal collaborative feature matrix. The status of casualties, infrastructure damage, and hazardous chemical risks constitute a pre-defined earthquake-specific identification framework, which defines the entire scope of independent propositions for disaster assessment.

[0110] Specifically, the earthquake emergency response terminal normalizes and assigns values ​​to the feature values ​​of the visual feature components and the semantic feature components according to the proposition classification criteria of the identification framework. It then converts the feature values ​​into the basic confidence proportions of each disaster proposition within the identification framework, completes the basic probability allocation conversion of the two types of feature components, and independently generates the basic probability allocation function for visual evidence and the basic probability allocation function for semantic evidence.

[0111] S32. Calculate the evidence distance conflict metric for the basic probability assignment functions of visual evidence and semantic evidence, and generate a distance metric value. The distance metric value is calculated using the following formula:

[0112]

[0113] in, For distance metrics, and These are the basic probability assignment functions for visual evidence and semantic evidence, respectively. This is the Jaccard distance matrix.

[0114] Optionally, the Jaccard distance matrix assigns predefined values ​​to its elements based on the overlap and difference between sets of different disaster propositions. The earthquake emergency response terminal transcribes the basic probability assignment function of visual evidence into a standard column vector form and denotes it as... The semantic evidence basic probability assignment function is rearranged into a standard column vector of the same dimension and denoted as . , The earthquake emergency response terminal first solves for a fixed Jaccard distance matrix pre-constructed based on the correlation relationships of the identification framework propositions. and The vector difference is then transposed and multiplied sequentially with the Jaccard distance matrix and the original vector difference. After the operation is completed, half of the overall result is taken and the square root is calculated. The distance metric that can characterize the degree of conflict between the two types of evidence is obtained by fully calculating according to the given formula.

[0115] S33. Perform a non-linear mapping of similarity and credibility weights on the distance metric to generate credibility allocation weights.

[0116] Optionally, the earthquake emergency processing terminal uses distance measurement values ​​as the input data source. The magnitude of the distance measurement value directly corresponds to the degree of conflict between the two pieces of evidence. The larger the distance measurement value, the lower the feature similarity between the visual evidence and the semantic evidence.

[0117] Specifically, the earthquake emergency processing terminal uses a continuous nonlinear transformation operation relationship to establish a reverse correspondence between distance measurement values ​​and feature similarity. First, it calculates the feature similarity between the basic probability allocation function of visual evidence and the basic probability allocation function of semantic evidence. Then, it substitutes the feature similarity into a preset nonlinear mapping rule for smooth transformation, converting the similarity value into a quantitative value that can be used for evidence weighting. Finally, it generates a credibility allocation weight that is suitable for the fusion operation of the two types of evidence.

[0118] S34. Based on the credibility-based weight allocation, perform local conflict reassignment on the basic probability allocation function of visual evidence and the basic probability allocation function of semantic evidence to generate the corrected visual evidence probability function and the corrected semantic evidence probability function.

[0119] Optionally, the earthquake emergency processing terminal uses the credibility allocation weight as the core matching basis to decompose and break down the conflicting parts in the basic probability allocation function of visual evidence and the basic probability allocation function of semantic evidence where there are discrepancies in proposition confidence judgment.

[0120] Specifically, the earthquake emergency processing terminal allocates the overall conflict confidence level to the disaster proposition dimension with higher confidence level according to the numerical proportion of the confidence weight. At the same time, it performs synchronous fine-tuning and correction on the confidence assignment of each proposition in the original basic probability allocation function to avoid the judgment error caused by the confidence bias of a single piece of evidence. It completes the local conflict redistribution processing between the two types of evidence and outputs the corrected visual evidence probability function and the corrected semantic evidence probability function after the assignment structure is corrected.

[0121] S35. Perform orthogonal summation on the modified visual evidence probability function and the modified semantic evidence probability function to generate a comprehensive probability quality function.

[0122] Optionally, the earthquake emergency processing terminal will uniformly convert the modified visual evidence probability function and the modified semantic evidence probability function into a probability vector form of the same dimension under the recognition framework. Orthogonal summation is an integration operation of the probability components of the same proposition in orthogonal space dimensions under the constraints of the evidence theory operation rules.

[0123] Specifically, for each independent disaster proposition—personnel casualties, infrastructure damage, and hazardous chemical risks—the earthquake emergency response terminal extracts the probability components corresponding to the modified visual evidence probability function and the modified semantic evidence probability function. It then performs component summation operations under orthogonal dimensions for each proposition, integrates and eliminates conflicting and redundant components, and unifies the integer value arrangement format to generate a comprehensive probability quality function that can centrally represent the confidence distribution of various disaster propositions.

[0124] In the above embodiments, the earthquake emergency response terminal achieves standardized probabilistic expression of visual and semantic evidence based on a standardized disaster identification framework through the probability allocation transformation of feature components to the disaster identification framework, evidence conflict measurement based on the Jaccard distance formula, nonlinear mapping of distance values ​​to confidence weights, redistribution of local evidence conflicts, and orthogonal summation of double-corrected evidence. It accurately quantifies the degree of evidence conflict through a given formula, reasonably resolves the judgment contradictions between heterogeneous evidence based on nonlinear mapping and weight ratio, effectively corrects the confidence bias of single-type evidence, and makes the final comprehensive probability quality function more objective and balanced in its probabilistic characterization of casualties, facility damage, and hazardous chemical risks.

[0125] In one embodiment, S4 may include:

[0126] S41. Obtain the pre-earthquake building vulnerability index, and perform structural failure risk quantification calculation on the probability of infrastructure damage state in the comprehensive probability quality function and the pre-earthquake building vulnerability index to generate a structural damage index.

[0127] Optionally, the building vulnerability index is a quantitative representation of the inherent susceptibility of different types of buildings to failure behaviors such as damage and collapse under earthquake action. The earthquake emergency response terminal retrieves the pre-stored regional pre-earthquake building vulnerability index and separately extracts the probability value corresponding to the infrastructure damage state from the comprehensive probability quality function. The earthquake emergency response terminal adopts disaster risk coupling calculation logic, assigning the probability of infrastructure damage state as a post-disaster external disaster-causing factor and the pre-earthquake building vulnerability index as a value of the building's own disaster resistance attribute. Through normalization correlation conversion, the two types of values ​​are linked and matched. By comprehensively considering the building's inherent damage resistance capacity and the probability of actual damage assessment after the earthquake, the failure risk at the building structure level is quantitatively solved, generating a structural damage index that can represent the overall severity of building damage.

[0128] S42. Obtain meteorological environmental parameters, perform quantitative calculations of hazardous chemical diffusion risk based on the probability of hazardous chemical risk state in the comprehensive probability mass function and meteorological environmental parameters, and generate a hazardous chemical risk index.

[0129] Optionally, meteorological environmental parameters serve as fundamental parameters characterizing atmospheric flow and environmental conditions in the disaster area, including conventional environmental parameters such as wind speed, wind direction, ambient temperature and humidity, and atmospheric stability. The earthquake emergency response terminal reads pre-stored real-time meteorological environmental parameters from the disaster area and extracts probability values ​​corresponding to the hazardous chemical risk status from the comprehensive probability mass function. Based on the conventional propagation correlation laws of hazardous chemical atmospheric diffusion, the terminal establishes a correspondence between meteorological environmental parameters and the spread trend of hazardous chemical leaks. It uses the probability of the hazardous chemical risk status as a basic confidence condition for the occurrence of a leak, and combines this with the constraints of meteorological environmental parameters on the diffusion range and rate to conduct linked quantitative calculations. This comprehensive assessment of the potential hazard level after a hazardous chemical leak generates a quantifiable hazardous chemical risk index that characterizes the disaster level.

[0130] S43. Extract the population time series baseline and evacuation time series rate from the unified spatiotemporal data cube. Based on the structural damage index and hazardous chemical risk index, perform risk quantification calculations on the trapped personnel baseline and damage amplification effect on the population time series baseline and evacuation time series rate to generate the casualty index. The formula for calculating the casualty index is as follows:

[0131]

[0132] in, The casualty index, for Population time series baseline at any given moment. for evacuation timing rate for Time-based structural damage index This is the damage amplification factor for hazardous chemicals to personnel. for Hazardous chemical risk index at all times.

[0133] Optionally, the population time series baseline is a representation of the baseline number of permanent and transient populations in the disaster area arranged in a time series within a unified spatiotemporal data cube, and the evacuation time series rate is a representation of the proportion of people in the disaster area who have been evacuated in an orderly manner at the corresponding time. The earthquake emergency processing terminal extracts the population time series baseline and evacuation time series rate from the unified spatiotemporal data cube time-series, and performs traversal calculations on each scale of the time dimension according to the given calculation formula. This represents the final, integrated casualty index. Represents any Population time series base at any given time Represents any The evacuation timing rate corresponding to each moment. Represents any The structural damage index corresponding to time . The fixed-set personal injury amplification factor for hazardous chemicals, Represents any The hazardous chemical risk index corresponds to a specific time. The earthquake emergency response terminal first calculates the base number of people trapped who have not been evacuated at a single time, then adds the weight of the personal threat from structural damage and the harm amplification effect of hazardous chemical risks, and sums up all the time-series calculation results to finally generate a casualty index that conforms to the time-series change pattern.

[0134] S44, combined structure damage index, hazardous chemical risk index and personnel casualty index, generate multi-dimensional quantitative assessment indicators.

[0135] Optionally, the earthquake emergency response terminal retrieves and calculates the structural damage index, hazardous chemical risk index, and casualty index, and performs numerical dimension unification and standardization processing on the three types of indices to ensure that the three types of indices maintain the same quantitative grading standards and numerical value ranges, thereby eliminating the dimensional differences between different assessment dimensions.

[0136] For example, the earthquake emergency response terminal integrates and collects the structural damage index (representing building safety status), the hazardous chemical risk index (representing environmental disaster status), and the casualty index (representing people's livelihood disaster status) in an orderly manner according to the classification and affiliation of disaster assessment. It does not change the original numerical representation meaning of each index, but only performs structured regularization and collection to form a multi-dimensional quantitative assessment indicator covering the three dimensions of infrastructure, hazardous chemical disasters, and people's disaster.

[0137] In the above embodiment, the earthquake emergency response terminal obtains a structural damage index by coupling building vulnerability attributes with the probability of post-disaster damage, and obtains a hazardous chemical risk index by combining meteorological conditions and the probability of hazardous chemical risks. Then, based on time-series population data and a given formula, it completes the time-series risk calculation of casualties. Finally, it integrates the three core indices to form a standardized assessment index. This embodiment integrates pre-earthquake inherent basic attributes, post-disaster multi-source determination probabilities, and time-series population change patterns, introduces the hazardous chemical damage amplification effect to fit the characteristics of the chain reaction of disasters, and uses fixed calculation rules to achieve a multi-dimensional quantitative characterization of the disaster situation, avoiding the shortcomings of traditional assessment methods that are single-dimensional and ignore time-series changes.

[0138] In one embodiment, S42 may include:

[0139] S421. Based on the probability of hazardous chemical risk status, perform abnormal smoke region segmentation on the visual data in the unified spatiotemporal data cube to generate abnormal smoke contour boundaries.

[0140] Optionally, smoke region segmentation is an image processing method that delineates the boundary of the target region based on the differences in image pixel features. The earthquake emergency processing terminal uses the probability of hazardous chemical risk status as a priori constraint to lock the key analysis range of visual data in a unified spatiotemporal data cube. The earthquake emergency processing terminal traverses the grayscale and texture features within the visual data pixel by pixel, classifies pixels based on the differences in pixel features between smoke regions and regular surface regions, and then removes isolated noise pixels and invalid fragmented regions through connected component filtering rules, retaining continuous and complete edge lines of the smoke-covered area. After smoothing and regularizing the edge lines, a closed range line structure is formed, ultimately generating an abnormal smoke contour boundary that can accurately delineate the actual smoke coverage area.

[0141] S422. Replace the abnormal smoke contour boundary with the default diffusion variance of the spatial concentration diffusion model to generate a dynamically corrected diffusion variance.

[0142] Optionally, the spatial concentration diffusion model is a fundamental operational paradigm for characterizing the natural diffusion and concentration gradient distribution of hazardous chemical gases in the atmosphere. The diffusion variance is used to quantitatively characterize the discrete extent of the gas diffusion coverage. The earthquake emergency response terminal retrieves the fixed default diffusion variance built into the spatial concentration diffusion model. This default diffusion variance is a preset fixed value adapted to general scenarios and lacks adaptability to real-world disaster areas. The earthquake emergency response terminal reads the actual spatial extension scale corresponding to the boundary of the abnormal smoke outline and directly replaces the original default diffusion variance value in the spatial concentration diffusion model with this real-world scale value. This achieves matching and correction between the diffusion parameters and the actual smoke spread range in the disaster area, generating a dynamically corrected diffusion variance that closely matches the actual diffusion situation on-site.

[0143] S423. Obtain the ground gas sensor concentration, perform leakage source strength inversion calculation on the dynamic correction diffusion variance and the ground gas sensor concentration, and generate dynamic leakage source strength.

[0144] Optionally, the ground gas sensor concentration is a real-time indicator of the on-site concentration of hazardous chemical gases collected by dedicated sensing equipment deployed in the disaster area, which can serve as a realistic observational constraint for diffusion simulation. The earthquake emergency response terminal reads the ground gas sensor concentration collected externally and integrated into a unified spatiotemporal data cube, uses the dynamically corrected diffusion variance as a boundary constraint for hazardous chemical diffusion, and employs numerical iterative solution logic to perform leakage source strength inversion calculation. The earthquake emergency response terminal uses the ground gas sensor concentration as a fitting reference standard, continuously iteratively adjusting the assigned value of the source emission intensity until the simulation-obtained concentration distribution and the ground gas sensor concentration converge. The intensity value locked after convergence is the dynamic leakage source strength that conforms to the actual on-site conditions.

[0145] S424. Obtain wind speed and direction from meteorological environmental parameters. Based on dynamic leakage source strength and dynamic correction diffusion variance, perform integral calculations on wind speed, wind direction, and spatial concentration parameters to generate a hazardous chemical risk index. The formula for calculating the hazardous chemical risk index is as follows:

[0146]

[0147] in, Indicates the risk index of hazardous chemicals. For dynamic leakage source strength, For wind speed, For the height of the leak, and To dynamically correct the components of diffusion variance in the horizontal and vertical directions, and For spatial coordinate variables, Let be the integral volume space.

[0148] Optionally, Gaussian diffusion distribution is a classic representation of the spatial distribution of pollutant concentrations in the atmosphere. The earthquake emergency response terminal extracts two core parameters—wind speed and wind direction—from meteorological environmental parameters, simultaneously determines the spatial concentration parameter distribution rules under a spatial coordinate system, and performs global integration calculations within the integration volume space according to a given integral formula. In the formula… To obtain the final hazardous chemical risk index, This represents the dynamic leakage source strength obtained from the preceding operations. These are the collected wind speed values. The height at which hazardous chemical leak sources are located. and These represent the horizontal and vertical components corresponding to the dynamically corrected diffusion variance, respectively. and These are the position variables in the spatial coordinate system. The defined three-dimensional integral volume space is used. The earthquake emergency response terminal follows the principle of atmospheric diffusion and mirror reflection to perform the superposition calculation of exponential terms, and then performs integration and accumulation over the entire integral volume space to finally generate a hazardous chemical risk index that can characterize the degree of hazard across the entire domain.

[0149] In the above embodiments, the earthquake emergency response terminal completes the smoke region segmentation of remote sensing visual data through the prior probability of hazardous chemical risks, obtains the dynamically corrected diffusion variance by correcting the inherent parameters of the diffusion model using the actual contour boundary of the smoke, obtains the dynamic leakage source strength by iterative inversion combined with ground gas sensor observation data, and then completes the full-domain risk quantification solution by integrating meteorological parameters such as wind speed and direction through the atmospheric diffusion integral formula. This embodiment combines remote sensing visual real-scene features, ground sensor measured data and classical atmospheric diffusion calculations, abandoning the extrapolation bias caused by fixed default parameters. It can dynamically adapt to the actual situation of hazardous chemical diffusion in the complex terrain and environment of earthquake-stricken areas, accurately depict the concentration distribution and hazard level of hazardous chemicals in three-dimensional space, and provide hazardous chemical risk quantification results that fit the actual situation on site for multi-dimensional disaster quantification assessment, effectively improving the accuracy and completeness of the assessment of secondary hazardous chemical disasters in the earthquake disaster chain.

[0150] In one embodiment, S5 may include:

[0151] S51. Classify the structural damage index, casualty index, and hazardous chemical risk index based on the natural breakpoints of the disaster chain ontology logic constraints, and generate logically consistent classification assessment indicators.

[0152] Optionally, the natural discontinuity grading automatically divides the grade intervals based on the data's own numerical distribution clustering characteristics, preserving the inherent clustering patterns of the data. The earthquake emergency response terminal retrieves the structural damage index, casualty index, and hazardous chemical risk index, and constrains the upper and lower limits of the grade intervals of each index according to the disaster chain ontology logic, ensuring that the grade division of different disaster indices conforms to the inherent induction and transmission patterns of the earthquake disaster chain. Following the numerical clustering rules of the natural discontinuity grading, the earthquake emergency response terminal performs interval merging and grade delineation for the global values ​​of each type of disaster index, while simultaneously correcting the boundary thresholds of the grade intervals according to the disaster chain ontology logic, avoiding logical conflicts in the grade standards of different disaster indices, ultimately generating a graded assessment index with well-defined grade divisions and unified disaster correlation logic.

[0153] S52. Dynamically symbolize and render logically consistent hierarchical assessment indicators to generate personnel casualty isosurface layers, hazardous chemical flow particle layers, and infrastructure damage base map layers.

[0154] Optionally, dynamic symbolic rendering is a spatial visualization method that matches specific graphic symbols and color gradients according to the disaster level. The earthquake emergency response terminal reads the spatial grid values ​​corresponding to the logically consistent hierarchical assessment indicators, and according to the geospatial grid mapping rules, uses isosurface interpolation to fit the hierarchical indicators related to casualties, connecting spatial points of the same disaster level to generate a casualty isosurface layer.

[0155] For example, the earthquake emergency response terminal uses a dynamic particle arrangement method to simulate the gas diffusion situation for hazardous chemical-related classification indicators, generating a hazardous chemical flow particle layer. At the same time, it matches the basic geographic base map texture and level color scheme for infrastructure damage classification indicators, generating an infrastructure damage base map layer.

[0156] S53. Based on the intensity of disaster indicators, dynamic transparency channel values ​​are assigned to the personnel casualty isosurface layer, hazardous chemical flow particle layer, and infrastructure damage base map layer for adaptive synthesis to generate an initial synthesized thematic map. The formula for calculating the fused pixel value of the initial synthesized thematic map is as follows:

[0157]

[0158] in, These are the output pixel values ​​of the initial composite thematic map. and The dynamic transparency of the personnel casualty layer and the hazardous materials layer are respectively. , and These are the original pixel values ​​corresponding to the personnel casualty layer, hazardous materials layer, and infrastructure layer, respectively.

[0159] Optionally, the transparency channel is used to control the pixel transparency ratio during the image layer overlay process, achieving a smooth blending transition of multiple image layers. The earthquake emergency processing terminal matches appropriate dynamic transparency channel values ​​to the personnel casualty isosurface layer and the hazardous chemical flow particle layer based on the intensity of the disaster index corresponding to each spatial location; the higher the intensity of the disaster index, the lower the transparency ratio of the corresponding layer. The earthquake emergency processing terminal performs pixel-by-pixel weighted calculations according to a given pixel fusion formula, where... These are the output pixel values ​​corresponding to the initial synthesized thematic map. The dynamic transparency corresponding to the casualty isosurface layer. This refers to the dynamic transparency of the hazardous chemical flow particle layer. , and The original pixel values ​​corresponding to the personnel casualty isosurface layer, the hazardous chemical flow particle layer, and the infrastructure damage base map layer are sequentially used to complete the adaptive pixel fusion of the three layers through nested weighted calculations, generating an initial composite thematic map.

[0160] S54. Calculate the inner product of structural damage gradient and hazardous chemical risk gradient for the spatial distribution of the infrastructure layer and hazardous chemical layer in the initial synthetic thematic map, and generate the indicator profile of the active area of ​​the disaster chain.

[0161] Optionally, the gradient numerical field characterizes the rate of increase or decrease of disaster indicators in the geographic spatial dimension, reflecting the spatial evolution trend of disaster intensity. The earthquake emergency processing terminal solves for the structural damage gradient numerical distribution of the infrastructure layer in the initial synthetic thematic map, and simultaneously solves for the hazardous chemical risk gradient numerical distribution in the hazardous chemical layer. It uses spatial vector inner product operation rules to perform coupling operations on the two types of gradient values ​​at the same geographic coordinate location. The gradient inner product operation result can characterize the degree of correlation and coupling between structural damage and hazardous chemical risk. The earthquake emergency processing terminal selects continuous spatial regions with gradient inner product values ​​higher than a conventional threshold, and smoothly outlines them along the region edges to generate a disaster chain active area indicator profile that can mark the high-incidence range of disaster chains.

[0162] S55. Overlay the active zone indicator outline of the disaster chain onto the initial synthetic thematic map to generate a comprehensive earthquake emergency thematic map.

[0163] Optionally, the earthquake emergency processing terminal extracts the vector boundary coordinate information corresponding to the active area indicator profile of the disaster chain, matches the active area indicator profile of the disaster chain to a unified geographic coordinate system, and aligns and overlays it to the top layer of the initial composite thematic map. The earthquake emergency processing terminal maintains the pixel blending effect and color rendering style of each layer within the initial composite thematic map without modification, only setting exclusive line styles and boundary labels for the active area indicator profile of the disaster chain, clearly distinguishing the spatial range of ordinary disaster areas and chain-linked active areas, supplementing the spatial distribution characteristics of the disaster chain while retaining the multi-dimensional disaster layer visualization information, and generating the final comprehensive earthquake emergency thematic map after integrating all layers and profile rendering content.

[0164] In the above embodiments, the earthquake emergency processing terminal uses three types of disaster indices to classify disasters based on the natural breakpoints of the disaster chain ontology logic. It then generates three differentiated disaster visualization layers through dynamic symbolic rendering. Next, it assigns transparency channel values ​​based on the intensity of disaster indicators to achieve adaptive pixel fusion of multiple layers, resulting in an initial thematic map. Following this, it extracts the active area indicator contours of the disaster chain through dual gradient inner product operations, and finally overlays the contours to complete the final thematic map generation. This embodiment uses the inherent logical constraints of the disaster chain to constrain the classification standard, relies on pixel-weighted fusion to achieve natural overlay of multiple layers, and uses gradient coupling to accurately locate high-incidence areas of disaster chains. This overcomes the shortcomings of traditional thematic maps, which only involve simple layer overlay and cannot reflect the correlation of disaster chains. The generated comprehensive earthquake emergency thematic map has clear layers, complete disaster dimensions, and clear spatial correlation characteristics, providing intuitive and reliable visualization support for earthquake emergency rescue deployment and overall disaster assessment.

[0165] In one embodiment, S54 may include:

[0166] S541. Perform spatial differential operator convolution on the pixel spatial distribution of the infrastructure layer and the hazardous chemicals layer in the initial synthesized thematic map to generate the spatial gradient field of structural damage and the spatial gradient field of hazardous chemicals risk.

[0167] Optionally, the spatial differential operator convolution is a conventional image processing operation that uses a fixed standard convolution kernel to solve for the rate of change of spatial values ​​of image pixels. The spatial differential operator convolution uses the Sobel gradient standard convolution kernel, which is a classic gradient detection kernel with a fixed size and fixed numerical arrangement. The earthquake emergency processing terminal uses the complete pixel arrays of the infrastructure layer and the complete pixel array of the hazardous chemicals layer in the initial synthesized thematic map as the operation carrier. The standard spatial differential operator convolution kernel is traversed through the neighboring pixels of the two layers in turn, and the rate of change of index values ​​in the horizontal and vertical spatial dimensions is calculated pixel by pixel. The rate of change of each pixel position is arranged globally to form a continuous global numerical distribution of the structural damage spatial gradient field and the hazardous chemicals risk spatial gradient field. Each pixel in the gradient field represents the degree of spatial change of the disaster index of the corresponding geographical location.

[0168] S542. Perform pixel-by-pixel multiplication and summation operations on the spatial gradient field of structural damage and the spatial gradient field of hazardous chemical risk to generate the gradient inner product field.

[0169] Optionally, the earthquake emergency processing terminal first completes the geographic coordinate pixel registration of the spatial gradient field of structural damage and the spatial gradient field of hazardous chemical risk, ensuring that the pixels in the same row and column of the two gradient fields correspond to completely consistent geographic spatial locations, and avoiding calculation deviations caused by spatial coordinate misalignment.

[0170] Specifically, the earthquake emergency processing terminal performs pixel-by-pixel vector multiplication on the two registered gradient fields, multiplies the gradient components at the same coordinate position, and then sums and normalizes the results of the local neighboring pixel multiplication. The terminal then reassigns values ​​according to the original spatial arrangement of the pixels to generate a new global numerical array, thus obtaining a gradient inner product field that can quantify the degree of correlation and coupling between the trend of structural damage and the trend of hazardous chemical risk. The pixel values ​​of the gradient inner product field directly correspond to the strength of the linkage between the two disaster gradients.

[0171] S543. Binarize the gradient inner product field with a set threshold and extract the edge to generate a disaster chain active area indicator profile.

[0172] Optionally, binarized edge extraction is a standard operation method that divides a continuous numerical field into two types of pixel attributes and delineates the boundary contours based on a fixed discrimination threshold. The earthquake emergency processing terminal calls the pre-set fixed discrimination threshold, traverses all pixel points inside the gradient inner product field, and classifies pixels with pixel values ​​greater than the set discrimination threshold as disaster-related valid pixels, and pixels with pixel values ​​less than the set discrimination threshold as ordinary background pixels. After completing the global binarization division, connected component aggregation is performed on discrete valid pixels to remove isolated and fragmented pixel units without actual spatial meaning. Then, the boundaries of continuous pixels are smoothed and fitted to delineate continuous and regular closed boundary lines, and finally, a disaster chain active area indicator profile that can accurately delineate the high-incidence range of disaster chain coupling is generated.

[0173] In the above embodiments, the earthquake emergency processing terminal sequentially solves the spatial gradient fields of the two types of disaster layers through convolution using standard spatial differential operators. Based on accurate pixel coordinate registration, it performs pixel-by-pixel multiplication and summation of the dual gradient fields to obtain the gradient inner product field. Then, it extracts the indicator profile of the active area of ​​the disaster chain based on fixed threshold binarization and edge contour fitting. This embodiment can accurately quantify the spatial gradient coupling relationship between infrastructure damage and hazardous chemical risks, clearly depict the spatial boundary of active disaster chain linkage, provide accurate vector contour basis for thematic map annotation rendering, and effectively improve the completeness and positioning accuracy of the earthquake emergency thematic map in expressing the spatial distribution characteristics of the disaster chain.

[0174] In the aforementioned method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, based on multi-source heterogeneous data including UAV remote sensing images, satellite images, ground sensor data, and rescue reporting text, a unified spatiotemporal data cube is first constructed by unifying the spatiotemporal benchmarks of various raw data. Then, layer-by-layer edge extraction of remote sensing images, guided filtering feature sharing, sensor and text data encoding mapping, and cross-attention operation are carried out to achieve cross-modal feature fusion of visual and semantic data. Evidence theory fusion is completed through evidence basic probability allocation, Jaccard distance conflict measurement, confidence weight nonlinear mapping, and local conflict redistribution to obtain a comprehensive probability quality function characterizing the confidence distribution of disaster. Subsequently, three assessment indices—structural damage, hazardous chemical risk, and casualties—are obtained by combining pre-earthquake building vulnerability, meteorological environmental conditions, and population temporal characteristics for hierarchical quantification. After disaster chain ontology logical hierarchical classification, multi-layer dynamic symbol rendering, and adaptive pixel synthesis, the active area of ​​the disaster chain is delineated based on spatial differential convolution, gradient field inner product operation, and binarized edge extraction. Finally, boundary profiles are superimposed to complete the generation of comprehensive earthquake emergency thematic maps. This technical solution enables deep correlation and fusion of multi-source heterogeneous disaster data at the feature level and resolves conflicts and uncertainties. It integrates the disaster chain coupling effect throughout the process to conduct multi-dimensional and hierarchical risk simulation, thus solving the technical defects of traditional technologies that simply overlay data, lack comprehensive consideration of disaster chains, have a single disaster assessment dimension, and cannot fully and accurately reflect the overall disaster situation.

[0175] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0176] Based on the same inventive concept, this application also provides an apparatus for implementing the above-mentioned method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data. The solution provided by this apparatus is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the apparatus for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data provided below can be found in the limitations of the method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data described above, and will not be repeated here.

[0177] In one exemplary embodiment, such as Figure 2 As shown, a device 10 for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data is provided, comprising:

[0178] The spatiotemporal unification module 11 can be used to acquire UAV remote sensing images, satellite images, ground sensor data and rescue reporting text data, and to unify the spatiotemporal reference of UAV remote sensing images, satellite images, ground sensor data and rescue reporting text data to generate a unified spatiotemporal data cube.

[0179] The cross-modal fusion module 12 can be used to perform cross-modal interactive fusion of visual data features and semantic data features in a unified spatiotemporal data cube to generate a cross-modal collaborative feature matrix;

[0180] Evidence fusion module 13 can be used to perform evidence theory fusion based on local conflict allocation on cross-modal collaborative feature matrices to generate a comprehensive probability quality function;

[0181] The cascaded assessment module 14 can be used to conduct a cascaded risk quantification assessment of the comprehensive probability quality function in combination with the disaster chain coupling effect, and generate multi-dimensional quantitative assessment indicators.

[0182] The thematic map generation module 15 can be used to perform dynamic layer synthesis based on disaster chain logical ontology and visual saliency of multi-dimensional quantitative assessment indicators to generate comprehensive earthquake emergency thematic maps.

[0183] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for generating a comprehensive earthquake emergency thematic map by fusing UAV remote sensing and multi-source data as described above.

[0184] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for generating an integrated emergency thematic map of earthquakes by fusing UAV remote sensing and multi-source data as described above.

[0185] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0186] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, characterized in that, The method includes: S1. Acquire UAV remote sensing images, satellite images, ground sensor data and rescue report text data, and unify the spatiotemporal reference of the UAV remote sensing images, satellite images, ground sensor data and rescue report text data to generate a unified spatiotemporal data cube. S2. Perform cross-modal interactive fusion on the visual data features and semantic data features in the unified spatiotemporal data cube to generate a cross-modal collaborative feature matrix; S3. Perform evidence theory fusion based on local conflict allocation on the cross-modal collaborative feature matrix to generate a comprehensive probability quality function; S4. Perform a cascaded risk quantification assessment on the comprehensive probability quality function, combining the disaster chain coupling effect, to generate multi-dimensional quantitative assessment indicators; S5. Dynamically synthesize the multi-dimensional quantitative evaluation indicators based on the disaster chain logical ontology and visual saliency to generate a comprehensive earthquake emergency thematic map.

2. The method according to claim 1, characterized in that, S2 includes: S21. Extract edge features from the UAV remote sensing images in the unified spatiotemporal data cube to generate a UAV edge feature map; S22. Based on the edge feature map of the UAV, guide filtering feature sharing is performed on the low-level features of the satellite image in the unified spatiotemporal data cube to generate a visual fusion feature map; S23. Encode and map the physical quantity statistical features of the ground sensor data and the word vector sequence of the rescue reporting text data in the unified spatiotemporal data cube to generate semantic feature vectors; S24. Using the visual fusion feature map as a query matrix and the semantic feature vector as a key matrix and value matrix, perform cross-attention calculation to generate a cross-modal attention score matrix; S25. Based on the cross-modal attention score matrix, perform weighted fusion mapping on the visual fusion feature map and the semantic feature vector to generate the cross-modal collaborative feature matrix, which includes visual feature components and semantic feature components.

3. The method according to claim 2, characterized in that, The S3 includes: S31. Map the visual feature components and the semantic feature components to the identification frameworks for personnel casualty status, infrastructure damage status and hazardous chemical risk status, respectively, and perform basic probability allocation transformation to generate basic probability allocation functions for visual evidence and basic probability allocation functions for semantic evidence. S32. Perform evidence distance conflict measurement on the basic probability allocation function of visual evidence and the basic probability allocation function of semantic evidence to generate a distance measurement value, which is calculated by the following formula: in, For distance metrics, and These are the basic probability assignment functions for visual evidence and semantic evidence, respectively. The Jaccard distance matrix; S33. Perform a non-linear mapping of similarity and credibility weights on the distance metric to generate credibility allocation weights; S34. Based on the credibility allocation weight, perform local conflict reassignment on the basic probability allocation function of visual evidence and the basic probability allocation function of semantic evidence to generate the corrected visual evidence probability function and the corrected semantic evidence probability function. S35. Perform orthogonal summation on the modified visual evidence probability function and the modified semantic evidence probability function to generate the comprehensive probability quality function.

4. The method according to claim 1, characterized in that, The S4 includes: S41. Obtain the pre-earthquake building vulnerability index, and perform structural failure risk quantification calculation on the infrastructure damage state probability in the comprehensive probability quality function and the pre-earthquake building vulnerability index to generate a structural damage index. S42. Obtain meteorological environmental parameters, perform hazardous chemical diffusion risk quantification calculation on the hazardous chemical risk state probability in the comprehensive probability mass function and the meteorological environmental parameters, and generate a hazardous chemical risk index. S43. Extract the population time series baseline and evacuation time series rate from the unified spatiotemporal data cube. Based on the structural damage index and the hazardous chemical risk index, perform risk quantification calculations on the trapped personnel baseline and the damage amplification effect on the population time series baseline and the evacuation time series rate to generate a casualty index. The calculation formula for the casualty index is as follows: in, The casualty index, for Population time series baseline at any given moment. for evacuation timing rate for Time-based structural damage index This is the damage amplification factor for hazardous chemicals to personnel. for The risk index of hazardous chemicals at all times; S44. Combine the structural damage index, the hazardous chemical risk index, and the personnel casualty index to generate the multi-dimensional quantitative assessment index.

5. The method according to claim 4, characterized in that, S42 includes: S421. Based on the probability of the hazardous chemical risk state, perform abnormal smoke region segmentation on the visual data in the unified spatiotemporal data cube to generate abnormal smoke contour boundaries. S422. Replace the abnormal smoke contour boundary with the default diffusion variance of the spatial concentration diffusion model to generate a dynamically corrected diffusion variance; S423. Obtain the ground gas sensor concentration, and perform leakage source strength inversion calculation on the dynamic corrected diffusion variance and the ground gas sensor concentration to generate dynamic leakage source strength. S424. Obtain wind speed and wind direction from meteorological environmental parameters. Based on the dynamic leakage source strength and the dynamic corrected diffusion variance, perform integral calculations on the wind speed, wind direction, and spatial concentration parameters to generate the hazardous chemical risk index. The calculation formula for the hazardous chemical risk index is as follows: in, This indicates the risk index of hazardous chemicals. The strength of the dynamic leakage source. For wind speed, For the height of the leak, and The components of the dynamically corrected diffusion variance in the horizontal and vertical directions are given. and For spatial coordinate variables, Let be the integral volume space.

6. The method according to claim 4, characterized in that, The S5 includes: S51. The structural damage index, casualty index and hazardous chemical risk index are classified into natural breakpoints based on the logic constraints of the disaster chain ontology, and logically consistent classification evaluation indicators are generated. S52. Dynamically symbolize and render the logically consistent hierarchical evaluation indicators to generate a personnel casualty isosurface layer, a hazardous chemical flow particle layer, and an infrastructure damage base map layer. S53. Based on the intensity of disaster indicators, dynamically allocate transparency channel values ​​to the personnel casualty isosurface layer, the hazardous chemical flow particle layer, and the infrastructure damage base map layer for adaptive synthesis to generate an initial synthesized thematic map. The formula for calculating the fused pixel value of the initial synthesized thematic map is as follows: in, The output pixel values ​​of the initial synthesized thematic map. and The dynamic transparency of the personnel casualty layer and the hazardous materials layer are respectively. , and These are the original pixel values ​​corresponding to the personnel casualty layer, the hazardous materials layer, and the infrastructure layer, respectively. S54. Calculate the inner product of structural damage gradient and hazardous chemical risk gradient for the spatial distribution of the infrastructure layer and hazardous chemical layer in the initial synthetic thematic map, and generate an indicator profile of the active area of ​​the disaster chain. S55. Overlay the active area indicator profile of the disaster chain onto the initial synthetic thematic map to generate the comprehensive earthquake emergency thematic map.

7. The method according to claim 6, characterized in that, S54 includes: S541. Perform spatial differential operator convolution on the pixel spatial distribution of the infrastructure layer and the hazardous chemicals layer in the initial synthesized thematic map to generate a spatial gradient field of structural damage and a spatial gradient field of hazardous chemicals risk. S542. Perform pixel-by-pixel multiplication and summation operations on the structural damage spatial gradient field and the hazardous chemical risk spatial gradient field to generate a gradient inner product field. S543. The gradient inner product field is binarized and edge extracted with a set threshold to generate the active area indicator profile of the disaster chain.

8. A device for generating comprehensive earthquake emergency thematic maps by fusing UAV remote sensing and multi-source data, characterized in that, The device includes: The spatiotemporal unification module is used to acquire UAV remote sensing images, satellite images, ground sensor data and rescue reporting text data, and to unify the spatiotemporal reference of the UAV remote sensing images, satellite images, ground sensor data and rescue reporting text data to generate a unified spatiotemporal data cube. The cross-modal fusion module is used to perform cross-modal interactive fusion of visual data features and semantic data features in the unified spatiotemporal data cube to generate a cross-modal collaborative feature matrix; The evidence fusion module is used to perform evidence theory fusion based on local conflict allocation on the cross-modal collaborative feature matrix to generate a comprehensive probability quality function; The cascaded assessment module is used to perform a cascaded risk quantification assessment of the comprehensive probability quality function in combination with the disaster chain coupling effect, and generate multi-dimensional quantitative assessment indicators. The thematic map generation module is used to perform dynamic layer synthesis based on the disaster chain logical ontology and visual saliency of the multi-dimensional quantitative assessment indicators to generate a comprehensive earthquake emergency thematic map.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images